Research Of Medical Images Fusion Technology Based On Multi-scale Analysis

Image fusion technology can make different modal medical images achieve the functions of information complementation and redundancy elimination.Among the commonly used image fusion technologies,multi-scale analysis methods that can obtain image features in multiple directions are the most commonly used.Multi-scale analysis methods can obtain image information of different scales and increase the accuracy of image description.However,the ability to extract subtle features in images is not strong,and the fusion rules need to be improved to fully extract image information and achieve better fusion effects.This dissertation takes multi-modal medical images as the research object,carries out the research of medical image fusion method based on multi-scale analysis,and realizes the fusion of brain structure images and functional images.main tasks as follows:1.Research the commonly used multi-scale analysis tools and summarize their advantages and disadvantages.Through the CT/MRI image fusion experiment,compare the fusion effects of five multi-scale analysis methods: pyramid transform,wavelet transform,discrete wavelet transform,non-subsampled contourlet transform,and non-subsampled shearlet transform,and analyze and find non-subsampled shearlet transform.Compared with other multi-scale analysis tools,the wave transform preserves the edge contour information of the source image more comprehensively,but its ability to capture the internal texture details and other features of the image is still insufficient.2.Under the framework of multi-scale analysis,low-frequency and high-frequency fusion rules are designed separately to improve the richness of detailed information of the image fusion results.A fusion algorithm combining pulse-coupled neural network and sparse representation is proposed.Pulse-coupled neural network is very effective in extracting detailed information such as image edge contours in high-frequency sub-bands.Sparse representation can better obtain image brightness information in low-frequency sub-bands.The algorithm used is compared with other five classic multi-scale analysis image fusion algorithms.The experimental results show that the image structure similarity,mutual information,edge information and other objective aspects are better,and the naked eye observation is clearer and brighter.3.In order to improve the problem of impulse coupled neural network with too many parameter settings and limited preservation of sparse representation details,an improved impulse coupled neural network and sparse representation algorithm are proposed,which use parameter adaptive pulse coupled neural network and convolution sparse representation fusion respectively.High frequency subband and low frequency subband.Experimental results show that the improved algorithm has improved information richness and image clarity,and achieved better fusion results.Performance indicators such as image structure similarity,mutual information,and edge information have increased.Let’s compare five medical image algorithms based on multi-scale analysis.From the subjective and objective evaluation,the fusion effect is better than other methods.This subject conducts experiments on grayscale images and color images,and compares them with multi-scale analysis algorithms based on pyramid transform,discrete wavelet transform,and non-subsampled contourlet transform.The subjective and objective analyses are carried out respectively.The experimental results show that this:the algorithm retains the edge contour information of the source image,effectively improves the quality of the fused image,and improves the richness of image details.